From: AAAI Technical Report FS-98-01. Compilation copyright © 1998, AAAI (www.aaai.org). All rights reserved. Towardsa Theory for a Sociable Software Architecture Emmett Davis (651) 699-4367 EmmettDa@aol.com University of Saint Thomas Graduate Programs in Software Mail # OSS301 2115 Summit Avenue Saint Paul, MN55105 Bonnie Holte Bennett (651) 962-5509 bhbennett@stthomas.edu University of Saint Thomas Graduate Programs in Software Mail # OSS301 2115 Summit Avenue Saint Paul, MN55105 in discourse communities; and 3. meaning from its intellectual dynamics over time. The representation of these provides the basis for advanced, intelligent link analysis. The next section describes the technological and commercial pressures on advanced information systems. Section III describes the theoretical tenants underlying our work. Section IV explores the benefits we see in pursuing this approach. Section V describes an approach to implementing this model. Section VI describes some mechanismsin such an implementation. Abstract Thecomplexrelationship of networksof informationthat exist on the weband in modem computersystemsrequires advancedlink-analysis capabilities in order to makesense of the deluge of data. Applications are emergingwhich use "intelligent agents"to gather data, however,they are rarely morethan specially tuned search engineswith the goal of gathering individual’s purchasing and chat proclivities. Wepropose a link-analysis architecture focusedon the individualwithattention on the individual’s "discourse communities."Webelieve this focus, along with the development and refmementof a discourse communitythesaurus can uniquely begin to modelthe social and cultural component’s of a user’s profile. This profile can help to beginto create moretruly intelligent agentsperformingricher link analysis. Motivation This section describes a number of technological and commercial factors which are putting extreme demands on advancedinformation systems. Business tempo is accelerating. Exponential information growth inundates us daily. Weexpect to interact with smart systems that learn our preferences and meet our specific needs. Wehope to "manage by wire" where we communicate our intentions and our agents "make it so." The "value chain" is identified as an abbreviated business cycle in which highly customized goodsarrive immediatelyas the consumerfeels the desire for the product (with, ideally no time in the wholesaler’s or the retailer’s hands, and thus, no inventory costs). As organizations grow, they seek the optimal balance between distributed operations and economiesof scale. Agents abound on the web, but most seem to exist to invade our privacy, not maintain is as all the predictions of the electronic "Jeeves" have predicted. All of these changes are dramatic and challenging for the information systems they use, and they provide the motivation for sociable software architecture. Introduction In order for advanced information systems to be truly intelligent, they will need a social and cultural context. Wewill need to understand and articulate the social and cultural environmentsof each information system. In the past information systems were isolated databases and the extent of our architectural theory was normalization. Then we were only plumbers linking pipes without regard to content or on the opposite extreme workingin only one domain. Future systems will need to interact with a variety of foreign systems. While object-oriented is a step forward in dealing with these challenges, we need to be clear about our philosophy relating to knowledgein a variety of types of systems ranging from databases to artificial intelligence and informationretrieval. Beyondobject oriented programming, information and knowledge require an explicit understanding of their cultural and social environmentfrom which existing data and information comes. We must articulate how information is used as knowledgeand howit is increased or destroyed. For each information system we must understand: 1. its scientific and academic discipline, industry, or discourse community;2. the roles of actors Exponential Information Growth Inundates Us. Tractability, the combinatoric impact of large numbersof interacting entities, increases the computationalresources needed to complete a task. "The challenge posed to businesses by the soon-to-be obscene amount of information is staggering. Yet the promise is fabulous as well, fueling hopes that knowledge management will finally come of age." (Bovet and Sheffi 1998). While Bovet and Sheffi are focusing on logistics, other examples, manyweb and agent related, easily come to Copyright ©1998,American Association for ArtificialIntelligence (x~av.aaai.org). Allrightsreserved. 27 mind. Our desires to create space stations, anticipate the weather, and predict geological cataclysms are instances of our need to understand "large numbersof interacting entities." all levels of organizations, industry, and society as a whole." (Gurbaxani and Whang1991). The effect is beginning of managementby wire of complex industries and value chains. Professionals will increasingly manage by instructing systems of their intention, without giving detailed orders. Smart Systems Become an Exponential Force Multiplier. (Davis and Botkin 1994) describe smart objects,"Products are smart because they filter and interpret information to enable the use to act more effectively .... Theyare interactive, they becomesmarter the more you use them, and they can be customized." (Davis and Botkin 1994). "The more you use knowledgebased offerings, the smarter the3, get." A knowledgebased system will learn more and put the newinformation to productive use. The Ritz-Carlton hotel uses a customer preference tracking system. All the hotels learn if a customer expresses a preference for six hypoallergenic pillows at one of the hotels. "The more you use knowledge-based offerings, the smarter Fou get." An example would be tutor systems that become group enhanced decision support systems. "Knowledge-based products and services adjust to changing circumstances." Not only do smart products tell you something is wrong, they also "tell you what to do." (Davis and Botkin 1994). David and Botkin recognize that, "knowledge-based businesses can customize their offerings." Smart systems will allow large organizations and alliances of organizations to ultra customize their services and products. Smart systems act like compoundinterest, an exponential force multiplier with growth. Value Chain: an Abbreviated Business Cycle. The value chain is a metaphor for all the processes and information involved fi’om acquiring raw material, creating products, and distributing them to customers." Only interacting organizations, which deploy information systems across supply chains, may survive. Increased integration within the supply chain will "create better information, which in turn should support lower inventories and improved financial performance. But the evolution of IT and diminishingtransaction costs will also lead to a fundamentalrestructuring of industry practices for distributing and supporting products .... The biggest impactof this restructuring will be on the intermediaries, such as distributors and warehouses,that handle physical goods and the information associated with them between the producer and final consumer." (Lewis and Talalayesky 1997). Organizational Behavior is Becoming Interorganizational. "The contemporary competitive environment requires interorganizational exchange and coordination of logistics information to achieve common goals .... The managerialand cultural aspects of logistics are as critical as the impact of IT. While the emergence of electronic data interchange (EDI) and electronic markets would seem to permit an increase in the numbers of suppliers firms can do business with, the opposite trend seems more evident: firms are developing close relationships with a small numberof major suppliers. For example, Wal-mart’s in-store inventories of items such as Tide and Crest are managed by Procter & Gamble, who have direct access to point-of-sale data." (Lewis and Talalayesky 1997) Suppliers manage inventories sometimeswithout any intervention by the retailer. In the retail grocery industry, the movementtoward "efficient consumer response" seeks to more precisely match inventory levels at the manufacturer, distributor and retailer to actual demand."Risks (such as transaction risks and agency costs) remain a real issue: how much information should be shared with suppliers? Should suppliers provide their customers with details of their manufacturing processes? What will happen if suppliers are allowed to see information on their competitors? A final issue is the developmentof performance indicators and reward mechanisms for managing partnerships between producers, distributors, retailers and transportation firms. For example, buyers and suppliers may need to decide how they will share reductions in Managementby Wire: Carry Out my Intentions. In aviation "flying by wire" is a term that indicates that computersystems are used to augmenta pilot’s ability to assimilate and react to rapidly changing environmental information. "By-wire" systems present selected abstractions of a few crucial environmental factors. Instrumentation and communicationtechnologies aid in evaluating alternative responses. Computer system intercepts the pilot’s commandand translates it into thousandsof detailed orders that orchestrate the systemin real time. Whenpilot’s fly by wire, they’re flying information representations of airplanes. The key is that the instructions are intentions, not orders. In a similar way"managingby wire" is the capacity to run a business by managing its informational representation. "Information systems have reduced decision information costs by allowing decision makerscost-effective access to information and powerful tools (e.g., simulation and econometric modeling) for analyzing the retrieved information. The improvementin decision quality in turn increases operational efficiency. For example, accurate forecasting of future demands, coupled with efficient handling of material flows and production scheduling, can achieve a significant reduction of inventory costs. Indeed the impact of this information revolution has been felt at 28 transportation and inventory costs. " Talalayesky 1997). (Lewis and Representation: Internal and External Internalize the Social, Externalize the Web Agency Theory. The reshaping of players in markets, including the demise of somefirms, means that people’s behavior will be changed. Information systems are crucial to this process. The alliances, which arise from value or supply chain management,must be supported by interorganizational information linkages. Thesealliances, as had electronic data interchange linkages, represent a relationship commitmentthat moves transactions away from the open market. These transactions require that organizations decide to implement interorganizational information technology. One example of the impact of information technology is in the size of organizations. According to Gurbaxani and Whang"firm size and the allocation of decision-making authority among the various actors in a firm are, to a considerable degree, determined by the costs associated with acquiring, storing, processing and disseminating information." (Gurbaxani and Whang 1991). Gurbaxani and Whang "argue that as decision-making rights are pushed downwardin the organizational pyramid, the costs of communicating information upward decrease while agency costs resulting from goal divergence increase. Therefore, decision rights in an organizational hierarchy should be located where the sum of these costs is minimized. Modem IT can reduce the costs of communicatinginformation by improving the quality and speed of information processing and management’s decision making, leading to more centralized management. At the same time, IT can also provide managementwith the ability to reduce agency costs through improved monitoring capabilities and performance evaluation schemes, inducing decentralization of decision making." (Gurbaxani and Whang1991). A rich debate about "representation" has swirled for decades. In the most recent chapter, RodneyBrooks has asserted that, "Representation is the wrong unit of abstraction in building the bulkiest parts of intelligent systems." (Brooks 1991). In another work, he states that in his robot creatures, "There are representations, or accumulationsof state, but these only refer to the internal workings of the system; they are meaningless without interaction with the outside world." (Brooks 1998). The only external representation or models of the world are those determined to be useful. Andthese representations are stored in the world, not in the agent. For sociable software architectures, accumulations of compresseddata and evaluations on its conclusions are internal representations of the system’s history. Meaningabout the "real" worldis stored in the world, in the history and projected interactions with humans. Meaningis derived only from interactions with real world. A side effect is that we are well prepared to support knowledgesystems. Knowledgeis potential action based on information, data with meaning. Aninsight we wish to contribute is that words (strings of characters and their vocal counterparts) are of the real world. Wetreat words as real objects, but not just as external representations of meaning. Wordsare data and not information. We must build meaning, create information, from local interactive processes with this data. While Brooks appropriately rejects internal representations (which would seem to include words) for robots, we treat them as sensor data, and the social interactions they indicate as the "real world". Is our culture and society any less real than gravity and physical objects? Artificial intelligence can "movearound" our culture as easily as it does whenBrooks’ creatures move aroundthe halls of his labs. For both of us actions are the crucible. Only words are part of the "outside world" just as sonar readings. Theoretoical Tennants for Sociable Software Architecture This section describes the theoretical tenants motivating our work. The issues discussed in the previous section indicate an important change in demands on technology. As information grows exponentially, as expectations grow to manage by wire, as the value chain tightens, as organizational behavior and information sharing increases, and as agency theory exerts its pressure in humanas well as algorithmic manifestations, there will be a need for software with an ability to interact in a richer social context. As we explore "sociable" software architecture our core concern is our concept and treatment of representation. What do the symbols we manipulate represent? How does one determine meaning? Is meaningstable over time? Discourse Communities Model the Social Realities of Interactions This observation has interesting implications when viewedin light of Hjorland’s observations in information science. Specifically, that "discourse communities" provide a tool for analysis, comprehension, and processingof social interactions. Following pragmatic objectivity, we hold that there is no intrinsic meaning for knowledge or information in isolation from action. Information’s meaning is neither subjective nor relative. Pragmatic objectivity is based on action theory/utility and a collective methodology. 29 Objective idealism will search subject matter in permanent,inherent characteristics of knowledge,or in permanent,inherent semantic relationships and try to establish standardized, permanent, fixed ways of analyzing relationships (disregarding their potential use). Subjective idealism, on the other hand, will search for the subject matter of a documentin either the author’s or the users’ subjective perception of the document and try to develop a theory of subject analysis based on either the author’s psychological world (as in classical hermeneutics) or the users’ psychological world (as is done in parts of modern cognitive viewpoint). on perceptions of these objects can learn to create useful behavior. "The key idea ... is to be using the world as its own model and continuously match the preconditions of each goal against the real world." (Brooks 1991). Semantics arise from the success or failure of actions based on outcomes of processes performed on the raw data. There is no intrinsic meaningor semantic content before this. There is no external representation except for "maps"and "behavior" in sociable software architecture. Social Interaction: Fromthe position of activity theory, we do not look on subject as either inherent characteristics or as something subjective in an individual way. Individual subjects’ behaviorin relation to use and to representation of information should be interpreted in the light of a (sub)disciplinary context. (Hjorland 1997). Meaningis social and collective, that is, weuse a methodological collectivism methodology. Weneed ... to see knowledge as a developed historicalcultural-social product. The alternative to studying individual subjects and individual informationseeking behavior is to study knowledge domains. That is to study their informational structures, their terminology, knowledge representation, communicationpatterns, all in connection with their theories of knowledge and theories of science. (Hjorland 1997) The discourse communityfor Hjorland’s information systems is an academic discipline. This generalizes to business and culture for systems that are more than information retrieval systems. The discourse community becomescentral. It is the sole source of meaning.Based on observing a discourse community’saction over time and by interacting with it, sociable software architectures can use and create information. Given a discourse community as our unit of work, and not just the individual, we becomeaware of the roles individuals have within the discourse community. Our approach must therefore be collectivist. The challenge is to implementa recording of cultural and social environment and behavior. If we can do this, we have a basis for intelligence. Again the key to intelligent meaning is action, not representation of a permanentworld. For Brooks’ creatures, there is intelligence without external representation. They operate in a real, sensory world with humans. Weassert that the Internet and its databases are also a real, sensory world with humans.It is a world filled with wordsand figures. Agentsoperating 3O Mentorship for Learning Because we need meaning for information and knowledge and meaning comes from interaction with humans, social interaction is vital to sociable software architecture. Social interaction can be structured in a variety of ways, but the most basic is that of mentorship. For elemental interactions with one human,systems should allow for a mentor. Sociable software architecture will need a mentor the way Star Trek’s Data needed Dr. Su. The mentor provides artificially structured or meaningful interactions. Brooks recognized this need when he used scaffolding and feedback for his creatures. Scaffolding and feedback allow humansand other creatures to learn from humans. The learning process allows us to share attention with the care giver. The care giver in return narrows our focus to what is relevant and most helpful in learning about the matter at hand. While sociable software architecture’ architecture will change its clustering or conclusions depending on human evaluation of the proposals, sociable software architecture has two other methodologies for mentors to use. Since sociable software architecture consists only of rules and data, a mentorcould directly intervene in one or the other. The use of Lenat’s Cyc, for example, is an addition that would improve sociable software architecture’ discernment of words. (Lenat and Guha1990). For sociable software architecture, there are many individuals in manydiscourse communities and there are roles for these individuals in their discourse communities. In either case, learning from humansis basic. Roles in the Discourse Community:Innovators, Trend Setters, Maintainers. Because meaning is not just psychological, but is also sociological, we need a social unit of meaning of work. The discourse community is that unit of meaning. While there are agents for individuals in sociable software architecture and the individual contributes to meaning (especially trendsetters), the individual is not the determiner of meaning. The meaning we seek to understand is not subjective at either the author’s or the consumer’send. Discourse communities though are composed of many individuals. Each individual mayhave one or more roles and these will change over time. The challenge will be to first place an individual in a discourse communityand then determine what roles they play. Someindividuals are innovators and early change agents. Other individuals are successful trend setters and while not the first to innovate, when they do so, their new direction will usually be successful. Other individuals are maintainers of the discourse community. These individuals provide the instruction needed by new membersto adapt to the existing discourse community. That an individual operates in several domainsor discourse communitiesis a source of richness, appropriately gray, however, knowledgeof an individual’s membershipin a discourse community and of their role in it is useful to the individual’s agent in predicting interest in newsites, for example, if I am a trendsetter in a Star Trek community and an increasing numberof innovators have been visiting and re~rning to "www.foo.com" then myagent is safe in directing meto it. Improving Problem Solving Capacity This section explores the benefits we see in pursuing this technical approach. An analysis of the potential improvementsto problem solving, consider the problems trying to f’md something on the web. Current web searches return thousands of hits to a query. Even meta searches are inadequate since the best they can do is showthe intersection of individual searches. Mysearch engine needs to know about me, what I do, what I’m interested in, and I’d like it to work overtime finding interesting sites, calling myattention to them, and not just selling menew things. I’d like it to learn about me, but to keep that information private, and not share it with vendors so they can bombard me with email, and slot me into the appropriate banner add categories. I’d like myweb search agent to be proactive, have good recall and f’md related work, and to guard my privacy. If it can learn by sharing information with other agents, fine, I’d like it to learn to become more customized, but all the while, it must safeguard my privacy. Changes Over Time are Important. Current web agents are either one time use, or require registration and its subsequent loss of privacy. Ideally, users would like persistent web agents under their control. Time is significant for discourse communities. There are several forms of time: online and offline; evolution of individual, community, individual in community; and world of words and of sensory data of the world of words. Two forms of time that are particularly important are the evolution of discourse community’s meaning and the interaction time with membersof a discourse community. The evolution of the discourse community’s meaning must be tracked. Paradigm shifts can sometimesrapidly occur. Changesin legislation or regulation, for example, can shift the meaning of what had been a stable representation of the world. Becausea sociable software architecture stores its representation of the world’s model in the world, sociable software architecture is both vulnerable to shifts and should be accurate in tracking them. The interaction time of the individual is entirely the business of the individual’s agent and is entirely personal. Understanding this sense of time will allow sociable software architectures to respond to queries in a "getback-to-you" modewhen queries require a bit of time. Also, sociable software architectures will need to understand whento present innovations or changes, that is, whento anticipate a query. Addedto all these processing times, we still have the time frames inherent in the raw source data. Therefore there is a need for data fusion and knowledge representation structures such as blackboards. While not significant in all situations, sociable software architectures will need to understand whether a new datumis a confirmation of a knownfact or a newfact. Anticipating Needs with Proactive Models. Many current information-seeking systems rely on a query before processing and returning solutions. Wepropose proactive solutions either triggered by a specific query or more often searching based on both past queries, probable queries, and queries of similar users. If the system designer considers the solution space as a social epistemic space, this becomes feasible. Information retrieval system, rather than searching for a subject string indexed to a solution, should consider Hjorland’s statement that: "The subject of a document is the epistemological potential of that document."(Hjorland 1997). Improving Recall and Relevance. Doing so will improveboth the recall and relevance and the timeliness of the solutions offered by systems. If we evaluate solutions using ecological behavior, we have a more robust evaluation system. "The understanding of an individual’s optimal choice in information seeking is based on some conditions about the ecosystem in which that individual forms a part." (Hjorland 1997).. If, for example, we evaluate the content of a representation on the basis of whetherit can contribute to the solution of a concrete problem or satisfy a need (such as one or more of Maslow’s hierarchy of needs), then we can search larger space continually, even before a specific query is asked. This would improve retrieval time of very large and distributed stores and processes. This allows for spontaneous pre-processing and continuous agent clustering of potential solution spaces. It permits use of local syntax and representation of potential solution 31 spaces. A sociable architecture also provides the robustness to handle diverse time frames and sources through data fusion. In order to handle the complexity and ultra-customization demanded by the people providing our motivation, we needed an implementation that was intrinsically distributed. Within this n-tiered system, universal structures and operations occur. Data compression, distributed processing, background processing, and data stores are resulting benefits. Maintaining Privacy. Unlike current systems, it is both possible and advantageous to implement privacy of discourse communityindividual members.Solution paths are through clusters’ of members’behavior separate from their identity. Perhaps if the current systems had been originally built by consumersinstead of vendors, there would be fewer added process and data store baggage of collecting individual identities in order to actively approach individuals to buy more services and products. Source Agents Source agents work with the raw source data whatever it maybe. In interactive systems this includes tracking the bulk data about use of source data by users. These source agents make the first approximation of sorting with demographics and other structural clues the source and use data into discourse communities. The source agents have discourse communitymaps and trees of their characteristics and overlap of these clusters. The source agents use a variety of tools including statistical, and node and link analysis. This analysis includes citation analysis whereappropriate against a tree of citations. "Citations" is used here in a very general sense. In addition to the usual bibliographic citations, we include hot links in html sources and even borrowed vocabulary or techniques, if they and their source can be identified. This identification and clustering essentially becomes input to one or more discourse community agents via blackboardsand other structures. Improving Efficiency with Distributed Processing. Amongthe pre-conditions of such social cognitive systems are a memory of similar others’ use and the ability to identify similar others. Advantageous in today’s networkedsystems wouldbe distributed stores, pre-processing and processing behaviors. Agents need to be able to evaluate user, vendor, and system data and behavior. Agents need to be able to seek information with or without a query. Customized to Individual Users’ Needs. Two postconditions are the capability to contact an individual in an appropriate style as chosenby the individual and to accept evaluation of the by the individual, both directly and by inference. Appropriate access includes language prerequisites and formats. Because of the need for sociable systems, manytypes of information systems must link with each other: databases and data warehousing systems; information retrieval applications; and artificial intelligence and decision support systems. Link analysis agents with intent controls own execution, and object oriented programming systems especially with polymorphism at trans-componentlevel are indications of this need. Also importantwill be the treatment of time. Implementation Discourse Community Agents Throughblackboards and other structures the agents use for a discourse communityare available to source agents. Both sets of agents share manyof the same resources, which are distributed so that discourse communityagents use the portions of the resources usually needed by them without drawing downthe whole resource pool. An example of this is the thesaurus. Certainly tools such as Lenat’s CYCwill provide a foundation, but the thesauri will grow through interaction betweensource and usage data. A more permanent thesaurus, or at least one with a longer historical memory, will be the metathesaurus where higher level patterns are stored. The thesauri will be trees within trees, a terminology semihypercube. The discourse communityagent thesauri are the keys to the compresseddata files created by source and other agents. The discourse communityagents and their data stores and thesauri are distributed by community. Some communities are themselves distributed, others are centralized. Model This section describes an approach to implementing a sociable software architecture. Sociable systems must interact with the primary users they serve. Often the primary users are humans. Humans learn and change the rules with which systems must stay abreast. Humansalso define the meaningof the reality or world sensible to systems. Often this sensible reality is data. The systems then have processes to use on the data, often transforming data into information and at times knowledge. Weuse an n-tier agent model, consisting of at least source, discourse community,and individual user agents. 32 Individual’s Agents Each individual served has an agent that understands that the individual is a member of several discourse communities and has different roles in each community. It is the partnership of the discourse community agent and the individual’s that generates the anticipatory responses from the individual’s agent. It is the individual’s agent that receives the queries and the evaluations to the responses to queries. The individual’s agent maintains the data identifiable to the individual. This agent is the one that creates and protects the individual’s privacy. This agent is also the maintainer of the individual’s personal thesaurus as well as the sense of appropriate time and formats for interacting with the individual. In an extreme example, this agent maycommunicatewith the individual solely through another person. Instances might include the parents of infants. Models of the Discourse Communities This section describes some of the mechanismsused to implement a sociable software architecture. Wedepend on similarity measuresor classification using vectors in an n-dimensional space. As source agents conjure a new discourse community or discourse community agents decide that a discourse community is splitting into separate communities, we create new agents for the new discourse community. The models of the discourse are multi-dimensional arrays. Vector analysis of the array dimensions provides us with a discourse community’scentroid or fingerprint characteristics. This includes the features of clusters of individuals performing a role within the discourse community. Lately we have been experimenting with compressing the data store by transformingsimilar data into layers of a tree. Thesewill be shared across agents as a source of the "meta knowledge"about the individual’s membershipin and roles related to discourse communities. Three benefits can be derived from this: data compression, facilitation of distributed processing, and background processing. The compression highlights similarities as well as outlier patterns. Wethen look for the rare patterns that over time becomemainstream. The individuals attached to the once rare patterns are the trendsetters. If their role persists, then we can view probable developmentsfor the community. Data Compression Derives Naturally from the Use of Discourse Community Thesauri All the agents regardless of level, create and use compressed data. Wecan compress data in such a way that it does not need to be uncompressedfor operations to occur. Inspired by wavelets, we maybe able to compress data so that the differences are highlighted, rather than obscured. A classic discourse community example is the collectors of Eighteenth Century Hong Kong stamps. Whena discourse communityis well defined, then the compression will be tighter. Essentially we create an array string of what is most similar. Wethen remove these dimensions from the arrays of all members and create a pointer to what is similar, or note differences with the similarity standard. The identifier array for Hong Kongstamps will overlap with those of stamp collectors and those interested in the history of HongKong. Thesauri terms and links most attached to the fingerprint array becomesthe key to the compression of source data. Weuse the fingerprint thesauri to store the similarities. This allows us to also restore the raw data to a specified level. This creating, dissembling, and reassembling initially will be common as the agents create hypotheses that they later abandon. Because the thesauri and the compressed data are actually the internal representation of the world, other agents and entities may use them as sociable software architecture evolves. Compartmentalization of Thesauri Facilitates Distributed and Background Processing The distributed architecture of sociable software architecture meansthat we must use techniques such as information fusion. Rawdata will arrive from a variety of sources in a variety of modes. Blackboardsand other such mechanismswill be used by sociable software architecture agents to communicate clearly on the details of the rawdata. This will be important for all the agents at every level, but especially true for the source agents whoare mining data streams. Conclusions The complexrelationship of networks of information that exist on the web and in moderncomputer systems requires advancedlink-analysis capabilities in order to makesense of the deluge of data. A numberof factors contribute to this including exponential information growth demandfor smart systems, desire to "manageby wire," the abbreviated value, and dynamicbusiness organizations. The agents currently available on the web 33 seem to have no memoryand exist only to exploit privacy. Wepropose a link-analysis architecture focused on the individual with attention on the individual’s "discourse communities."Webelieve this focus, along with the developmentand ref’mementof a discourse communitythesaurus can uniquely begin to modelthe social and cultural component’sof a user’s profile. This profile can help to begin to create moretruly intelligent agents. The goals of this paper have been to begin to delineate the pressures towards more advanced intelligent agents which will provide a richer frameworkfor link analysis, to survey somerelated work from information theory, and to suggest an architecture which might be useful. We suspect that we’ve opened more questions than we’ve answered, and (an admirable thing to do in some respects), and we hope that our further research will continueto address these pertinent issues. References Bovet, D. and Sheffi, Y. "The Brave NewWorld of Supply Chain Management." Supply Chain Management Review. Vol 2. No. 1 Spring 1998. 14-22. Brooks, A. Rodney,et al.. "Alternative Essences of Intelligence" in Proceedings:Fifteenth National Conferenceon Artificial Intelligence (AAA1-98)and Tenth Conferenceon Innovative Applications of Artificial Intelligence (IAAI-98) 961-968. MenloPark, CA: AAAI Press/MITPress, 1998. Brooks, R., "’Intelligence without representation," Artificiallntelligence, vol. 47, 139--160,Jan. 1991. Byrne, S. M. and Javad, S. "Integrated Logistics Information Systems (ILIS): Competitive Advantage Increased Cost?" Annual Conference Proceedings (Oak Brook, IL: Council of Logistics Management,1992). 5573. Davis, S. and Botkin, J. "The Comingof KnowledgeBased Business." Harvard Business Review. Sept-Oct 1994. 165-170. Gurbaxani, V. and Whang,S. "The Impact of Information Systems on Organizations and Markets." Communications oftheACM. Vol. 34. No. 1. January 1991. 59-73. Hjorland, B. Information Seeking and Subject Representation: An ActiviO,-Theoretical Approachto Information Science. Westport, CT: GreenwoodPress, 1997. ISBN: 0-313-29893-9 34 Lenat, D.B. and Guha, R.V. "Building Large Knowledge-BasedSystems", Addison-Wesley, Reading, MA,1990. ISBN0-201-51752-3. Lewis, I. and Talayevsky, A.. "Logistics and Information Technology: A Coordination Perspective." Journal of Business Logistics. Vol. 18. No. 1. 1997. 141-157.